Julia Kempe

OpenApps: Simulating Environment Variations to Measure UI Agent Reliability

How reinforcement learning after next-token prediction facilitates learning

Soft Tokens, Hard Truths

From Concepts to Components: Concept-Agnostic Attention Module Discovery in Transformers

Embedding Trust: Semantic Isotropy Predicts Nonfactuality in Long-Form Text Generation

Don’t Waste Mistakes: Leveraging Negative RL-Groups via Confidence Reweighting

Tuning without Peeking: Provable Privacy and Generalization Bounds for LLM Post-Training

What Characterizes Effective Reasoning? Revisiting Length, Review, and Structure of CoT

Outcome-based Exploration for LLM Reasoning

Universal Likelihood Rewards for LLM Reasoning

Strong Model Collapse

DRoP: Distributionally Robust Data Pruning

On the Geometry of Regularization in Adversarial Training: High-Dimensional Asymptotics and Generalization Bounds

Beyond Model Collapse: Scaling Up with Synthesized Data Requires Verification

Flavors of Margin: Implicit Bias of Steepest Descent in Homogeneous Neural Networks

Asymmetric REINFORCE for off-Policy Reinforcement Learning: Balancing positive and negative rewards

Emergent properties with repeated examples

PILAF: Optimal Human Preference Sampling for Reward Modeling